U.S. patent number 11,339,763 [Application Number 15/828,450] was granted by the patent office on 2022-05-24 for method for windmill farm monitoring.
This patent grant is currently assigned to Hitachi Energy Switzerland AG. The grantee listed for this patent is Hitachi Energy Switzerland AG. Invention is credited to Ni Ya Chen, Yao Chen, Moncef Chioua, RongRong Yu, Yingya Zhou.
United States Patent |
11,339,763 |
Chioua , et al. |
May 24, 2022 |
Method for windmill farm monitoring
Abstract
A method for monitoring turbines of a windmill farm includes:
providing a global nominal dataset containing frame data of the
turbines of the windmill farm and continuous reference monitoring
data of the turbines for a first period in a fault free state, the
reference monitoring data including at least two same monitoring
variables for each turbine; building a nominal global model based
on the global nominal dataset which describes the relationship in
between the windmill turbines and clustering the turbines according
thereto; assigning the data of the global nominal dataset to
respective nominal local datasets according to the clustering; and
building a nominal local model for the turbines of each cluster
based on the respective assigned nominal local datasets, the
nominal local model being built such that a nonconformity index is
providable which indicates a degree of nonconformity between data
projected on the local model and the model itself.
Inventors: |
Chioua; Moncef (Heidelberg,
DE), Chen; Ni Ya (Beijing, CN), Yu;
RongRong (Beijing, CN), Zhou; Yingya (Shanghai,
CN), Chen; Yao (Beijing, CN) |
Applicant: |
Name |
City |
State |
Country |
Type |
Hitachi Energy Switzerland AG |
Baden |
N/A |
CH |
|
|
Assignee: |
Hitachi Energy Switzerland AG
(Baden, CH)
|
Family
ID: |
1000006326684 |
Appl.
No.: |
15/828,450 |
Filed: |
December 1, 2017 |
Prior Publication Data
|
|
|
|
Document
Identifier |
Publication Date |
|
US 20180087489 A1 |
Mar 29, 2018 |
|
Related U.S. Patent Documents
|
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
Issue Date |
|
|
PCT/EP2015/062389 |
Jun 3, 2015 |
|
|
|
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
F03D
17/00 (20160501); G06F 11/30 (20130101); Y02B
10/30 (20130101); F05B 2240/96 (20130101) |
Current International
Class: |
G06G
7/48 (20060101); G06F 11/30 (20060101); F03D
17/00 (20160101) |
Field of
Search: |
;703/6,18 |
References Cited
[Referenced By]
U.S. Patent Documents
Foreign Patent Documents
|
|
|
|
|
|
|
WO 2009016020 |
|
Feb 2009 |
|
WO |
|
Other References
Zhang et al. (Wind turbine fault detection based on SCADA data
analysis using ANN, 10 pages). (Year: 2014). cited by examiner
.
Kim et al. (A Three Dimensional Clustering in Wind Farms with
Storage for Reliability Analysis, 6 pages (Year: 2013). cited by
examiner .
Kim Hagkwen et al.: "Three dimensional clustering in wind farms
with storage for reliability analysis", 2013 IEEE Grenoble
Conference, IEEE, Jun. 16, 2013 (Jun. 16, 2013), pp. 1-6,
XP032519762. cited by applicant.
|
Primary Examiner: Louis; Andre Pierre
Attorney, Agent or Firm: Slater Matsil, LLP
Parent Case Text
CROSS-REFERENCE TO PRIOR APPLICATIONS
This application is a continuation application of International
Application No. PCT/EP2015/062389, filed on Jun. 3, 2015. The
entire disclosure of that application is hereby incorporated by
reference herein.
Claims
What is claimed is:
1. A method for monitoring turbines of a windmill farm, the method
comprising: providing a global nominal dataset containing frame
data of the turbines of the windmill farm and continuous reference
monitoring data of the turbines for a first period in a fault free
state, the reference monitoring data comprising at least two same
monitoring variables for each turbine; building a nominal global
model based on the global nominal dataset and relationships between
the windmill turbines present in the windmill farm by statistical
techniques during a fault-free time period, and clustering the
turbines according thereto; assigning the data of the global
nominal dataset to respective nominal local datasets according to
the clustering; building a nominal local model for the turbines of
each cluster based on the respective nominal local datasets and
based on multivariate statistical algorithms or artificial
intelligence techniques, wherein the nominal local model is built
in that way, that a nonconformity index (NC) is providable which is
indicating a degree of nonconformity between data and the model;
providing a test dataset with continuous test monitoring data of
the turbines of the windmill farm for a further period, wherein
those continuous test monitoring data are structured in the same
way than the continuous reference monitoring data in the nominal
global dataset and wherein the clustering of the nominal global
dataset is also applied on the test dataset; cluster wise
projection of continuous test monitoring data of the test dataset
on the respective assigned nominal local models of the turbines and
deriving a nonconformity index (NC) for each respective turbine
therefrom; and indicating a first turbine as critical in case that
the respective related nonconformity index exceeds a given
limit.
2. The method for monitoring turbines of a windmill farm of claim
1, wherein the nominal local model is based on Principal Component
Analysis, Linear Discriminant Analysis, Support Vector Machines, or
artificial intelligence techniques.
3. The method for monitoring turbines of a windmill farm of claim
1, further comprising refining the nominal local model during one
or more iterations, each iteration comprising: identifying one or
more turbines of the cluster of turbines as outliers; and
rebuilding the nominal local model without data collected from the
one or more turbines of the cluster of turbines identified as
outliers.
4. The method for monitoring turbines of a windmill farm of claim
3, wherein data collected from the one or more turbines of the
cluster of turbines identified as outliers is removed from the
subset of the global nominal dataset during each iteration.
5. The method for monitoring turbines of a windmill farm of claim
1, wherein the global nominal dataset comprises data for each
turbine of the turbines of the windmill farm comprising electrical
measurements, temperature measurements, motional measurements, or
ambient condition measurements.
6. The method for monitoring turbines of a windmill farm of claim
1, wherein the global nominal dataset comprises data about a type
of each turbine of the turbines of the windmill farm or a spatial
proximity of each turbine of the turbines of the windmill farm to
each other turbine of the turbines of the windmill farm.
7. The method for monitoring turbines of a windmill farm of claim
1, wherein each of the global nominal dataset and the test dataset
comprises a three mode dataset comprising several process variables
(index J) of several turbines (index I) along several time samples
(index K).
8. The method for monitoring turbines of a windmill farm of claim
1, wherein the global nominal dataset or the test dataset are
collected at least predominantly by a SCADA system.
9. The method for monitoring turbines of a windmill farm of claim
1, further comprising a computing device with a respective software
program module running on the computing device configured to
automatically perform the method.
10. The method for monitoring turbines of a windmill farm of claim
1, wherein an automatic fault analysis is initiated upon
identifying the first turbine as critical.
11. The method of claim 10, wherein the automatic fault analysis
comprises: turbine level parsing; time level parsing; and variable
level parsing.
12. The method for monitoring turbines of a windmill farm of claim
2, wherein the artificial intelligence techniques comprise a neural
network.
13. The method for monitoring turbines of a windmill farm of claim
5, wherein the electrical measurements comprise generated
electrical power measurements, voltage measurements, current
measurements, or power factor measurements, wherein the temperature
measurements comprise nacelle temperature measurements or
electrical generator temperature measurements, wherein the motional
measurements comprise blade speed measurements, or electrical
generator speed measurements, or wherein the ambient condition
measurements comprise wind direction measurements, wind speed
measurements, or ambient temperature measurements.
14. A method for monitoring turbines of a windmill farm, the method
comprising: collecting a global nominal dataset from a plurality of
turbines of a windmill farm during a reference period that is
determined to be fault free; building a nominal global model based
on the global nominal dataset and a plurality of relationships
between the plurality of turbines present in the windmill farm by
statistical techniques during a fault-free time period; identifying
a cluster of turbines from the plurality of turbines based on the
nominal global model; building a nominal local model for the
cluster of turbines based on a subset of the global nominal dataset
comprising data collected from the cluster of turbines during the
reference period; iteratively refining the nominal local model,
each iteration comprising: identifying an outlier turbine of the
cluster of turbines as being an outlier based on results obtained
from the nominal local model; removing the outlier turbine from the
cluster of turbines; removing data collected from the outlier
turbine from the subset of the global nominal dataset; and
rebuilding the nominal local model based on the subset of the
global nominal dataset after removing data collected from the
outlier turbine from the subset of the global nominal dataset;
collecting a test dataset from the cluster of turbines during an
operational period of the windmill farm; deriving a nonconformity
index for each turbine of the cluster of turbines that measures
conformity between the test dataset and the nominal local model;
and identifying a first turbine of the cluster of turbines as
critical when the nonconformity index for the turbine of the
cluster of turbines exceeds a given limit.
15. The method for monitoring turbines of a windmill farm of claim
14, wherein each of the global nominal dataset and the test dataset
comprises a three mode dataset comprising several process variables
(index J) of several turbines (index I) along several time samples
(index K).
16. The method for monitoring turbines of a windmill farm of claim
14, wherein the global nominal dataset or the test dataset are
collected at least predominantly by a SCADA system.
17. The method for monitoring turbines of a windmill farm of claim
14, further comprising continuing to refine the local nominal model
until no outlier turbines can be identified in the cluster of
turbines.
18. A method for determining a cluster of turbines of a windmill
farm, the method comprising: collecting a global nominal dataset
from a plurality of turbines of a windmill farm during a reference
period that is determined to be fault free; building a nominal
global model based on the global nominal dataset and a plurality of
relationships between the plurality of turbines present in the
windmill farm by statistical techniques during a fault-free time
period; identifying a cluster of turbines from the plurality of
turbines based on the nominal global model; building a nominal
local model for the cluster of turbines based on a subset of the
global nominal dataset comprising data collected from the cluster
of turbines during the reference period; and iteratively narrowing
the cluster of turbines by: identifying outlier turbines based on
results obtained from the nominal local model, removing outlier
turbines from the cluster of turbines, removing data collected from
the outlier turbines during the reference period from the subset of
the global nominal dataset, and rebuilding the nominal local model
based on the subset of the global nominal dataset after removing
data collected from the outlier turbines from the subset of the
global nominal dataset.
19. The method for determining a cluster of turbines of a windmill
farm of claim 18, wherein the global nominal dataset is collected
at least predominantly by a SCADA system.
20. The method for determining a cluster of turbines of a windmill
farm of claim 18, wherein the global nominal dataset comprises data
for each turbine of the plurality of turbines comprising electrical
measurements, temperature measurements, motional measurements, or
measurements describing ambient conditions.
Description
FIELD
The invention is related to a method for monitoring turbines of a
windmill farm. It is known, that Wind energy is currently the
fastest growing source of electric generation in the world.
Operation and maintenance, including scheduled and unscheduled
maintenance typically amounts 20% to 25% of the total windmill farm
project effort. Continuously monitoring the condition of windmill
turbines is seen as the most efficient way to reduce maintenance
effort of windmill turbines in that continuous monitoring with
integrated fault detection allow for early warnings of mechanical
and electrical faults to avoid unscheduled maintenance and
unnecessary scheduled maintenance.
BACKGROUND
Typically a Condition Monitoring System (CMS) is foreseen to
evaluate the condition of the components in a system such as a
windmill turbine. Fault detection is a Boolean decision about the
existence of faults in a system. The goal of a fault diagnosis is
the determination of the exact location and magnitude of a fault.
To date, several windmill turbine CMSs are available on the market
and many windmill turbine condition monitoring schemes have been
proposed in literature. These schemes can be classified according
to three aspects:
CMS can be implemented for a single component, a single turbine or
a set of multiple turbines. While extensive investigations have
been made in the area of single component monitoring such as e.g.
gearbox monitoring and in the area of single windmill turbine
monitoring according to its performance only few approaches exist
in monitoring multiple turbines using a single model, in particular
by obtaining positive results of monitoring multiple turbines by
tracking their relationship, it could achieve fault detection but
no fault diagnosis since it uses the measured power generation
variable as the only variable monitored for each turbine and
included in the model.
A model of windmill turbines and their components can be obtained
based on physical laws, using neural networks or statistical data
mining techniques. Modeling using statistical methods is often less
costly than modeling based on physical laws and leads to an easier
interpretability when compared to modeling using neural
networks.
Windmill turbine data can be collected from Supervisory Control and
Data Acquisition (SCADA) systems. SCADA systems are primarily used
for operating and controlling windmill turbines. Windmill turbine
data can be generated from additional installed sensors
specifically designed for CMS. Using SCADA data for condition
monitoring is motivated by the fact that data are readily
collected, requiring therefore no additional equipment engineering,
installation and testing.
Disadvantageously within the state of the art is that most of the
available condition monitoring or fault diagnosis systems are
focused a single windmill turbine, where the objective is to detect
whether a fault happens in the turbine. Such a turbine focused
approach is subject to a certain inaccuracy and also forthcoming
faults are not easily to detect since only information which are
directly related to the turbine are used for decision making.
SUMMARY
In an embodiment, the present invention provides a method for
monitoring turbines of a windmill farm, comprising the following
steps: providing a global nominal dataset containing frame data of
the turbines of the windmill farm and continuous reference
monitoring data of the turbines for a first period in a fault free
state, the reference monitoring data comprising at least two same
monitoring variables for each turbine; building a nominal global
model based on the global nominal dataset which describes the
relationship in between the windmill turbines and clustering the
turbines according thereto; assigning the data of the global
nominal dataset to respective nominal local datasets according to
the clustering; building a nominal local model for the turbines of
each cluster based on the respective assigned nominal local
datasets, the nominal local model being built such that a
nonconformity index is providable which indicates a degree of
nonconformity between data projected on the local model and the
model itself; providing a test dataset with continuous test
monitoring data of the turbines of the windmill farm for a further
period, those continuous test monitoring data being structured in a
same way as the continuous reference monitoring data in the nominal
global dataset, the clustering of the nominal global dataset being
also applied on the test dataset; cluster wise projection of
continuous test monitoring data of the test dataset on the
respective assigned nominal local models of the turbines and
deriving a nonconformity index for each respective turbine
therefrom; and indicating a turbine as critical when the respective
related nonconformity index exceeds a given limit.
BRIEF DESCRIPTION OF THE DRAWINGS
The present invention will be described in even greater detail
below based on the exemplary figures. The invention is not limited
to the exemplary embodiments. Other features and advantages of
various embodiments of the present invention will become apparent
by reading the following detailed description with reference to the
attached drawings which illustrate the following:
FIG. 1 shows the exemplary steps of the proposed workflow,
FIG. 2 shows an exemplary nominal global model cluster,
FIG. 3 shows an exemplary flow chart of the "nominal global model
building" step,
FIG. 4 shows an exemplary flow chart of the "nominal local model
building" step,
FIG. 5 shows an exemplary flow chart of the "test data projection"
step,
FIG. 6 shows an exemplary fault diagnostics on test monitoring
data,
FIG. 7 shows an exemplary turbine level parsing (level I),
FIG. 8 shows an exemplary time level parsing (level II),
FIG. 9 shows an exemplary variable level parsing (level III):
faulty variable isolation using statistical confidence limit,
FIG. 10 shows an exemplary variable level parsing (level III):
faulty variable isolation using comparison to contribution during
nominal period,
FIG. 11 shows an exemplary signal level display (level IV), and
FIG. 12 shows a schematic representation of a method for monitoring
turbines of a windmill farm.
DETAILED DESCRIPTION
The problem is solved by a method for monitoring turbines of a
windmill farm. This is characterized by the following steps:
providing a global nominal dataset containing frame data of the
turbines (122, 124) of the windmill farm (120) and continuous
reference monitoring data of the turbines (122, 124) for a first
period in the fault free state, wherein the reference monitoring
data comprise at least two same monitoring variables for each
turbine (122, 124),
building a nominal global model based on the global nominal dataset
which describes the relationship inbetween the windmill turbines
and clustering the turbines according thereto,
assigning the data of the global nominal dataset to respective
nominal local datasets according to the clustering,
building a nominal local model for the turbines of each cluster
based on the respective assigned nominal local datasets, wherein
the nominal local model is built in that way, that a nonconformity
index (NC) is providable which is indicating the degree of
nonconformity between data projected on the local model and the
model itself,
providing a test dataset with continuous test monitoring data of
the turbines of the windmill farm for a further period, wherein
those continuous test monitoring data are structured in the same
way than the continuous reference monitoring data in the nominal
global dataset and wherein the clustering of the nominal global
dataset is also applied on the test dataset,
cluster wise projection of continuous test monitoring data of the
test dataset on the respective assigned nominal local models of the
turbines and deriving a nonconformity index (NC) for each
respective turbine therefrom,
indicating a turbine as critical in case that the respective
related nonconformity index exceeds a given limit.
Basic idea of the invention is to take a holistic view of the whole
windmill farm and to use the similarity between the expected
behaviors of a subset of windmill turbines to determine whether or
not some windmill turbines exhibit abnormalities in their
behavior.
The algorithm used to model the nominal global and/or local model
respectively the relationship between windmill turbines can be but
is not limited to multivariate statistical algorithms such as
Principal Component Analysis, Linear Discriminant Analysis and
Support Vector Machines, or artificial intelligence techniques such
as neural network.
Depending on the type of algorithm used, one or more indices may be
developed to indicate the degree of nonconformity (denoted NC index
in the sequel) between data and model. The NC index together with
its statistical confidence limit is used to check:
The similarity of windmill turbines in the same model;
The dissimilarity of one or several windmill turbine(s) to other
turbines in the same model;
The nonconformity of one or several windmill turbine(s) during a
given time interval of operation.
Historical operational data are preferably collected from SCADA
system during periods where windmill turbines are fault-free and/or
operate in acceptable conditions. These periods form a nominal
operating condition dataset respectively the global nominal dataset
that is used as a reference for monitoring the windmill farm.
Data collected during those periods when the condition of the
windmill turbine is to be monitored and diagnosed is taken as base
of the test dataset. Both nominal data and test data are organized
in the same structure, recording the same variables of the same
turbines in the same windmill farm but during different time
periods. The dataset might be preferably in essence a three mode
dataset comprising several process variables (index J) of several
turbines (index I) measured along several time samples (index
K).
The variables can be for example signals related to the operation
of a windmill turbine, such as electrical measurements (e.g.
generated electrical power, voltage, current, power factor . . . ),
temperature measurements (e.g. nacelle temperature, electrical
generator temperature . . . ) and motional measurements (e.g. blade
speed, electrical generator speed . . . ) as well as measurements
variables describing the ambient conditions (e.g. wind direction,
wind speed and ambient temperature).
The described invention is related to a method to monitor windmill
farm solely based on historical data readily available for example
on a SCADA system. This is providing the following advantages:
Windmill turbines presenting abnormalities in their operation are
directly determined. This information might be presented
automatically to an operator so that his reaction time for starting
counteraction is reduced in an advantageous way. Of course it is
also possible to start counteractions automatically by the
monitoring system itself
Automatic root cause analysis in case of the occurrence of a
windmill turbine abnormal operation situation is as well enabled as
assisting an operator in root cause analysis.
Extensive high performance hardware and models are not required in
an advantageous way, since the method of the invention is a purely
data driven approach which is based on already existing data from
SCADA systems for example.
According to a further embodiment of the invention the local model
for the turbines of each cluster is based on multivariate
statistical algorithms such as Principal Component Analysis, Linear
Discriminant Analysis and Support Vector Machines or artificial
intelligence techniques such as neural network. Such methods, in
particular the statistical based methods, are easily implementable
and applicable on an existing database.
According to a further embodiment of the invention the nominal
local model for the turbines of each cluster is built iteratively,
wherein the data of those turbines which are not matching into the
local model are identified as outliers and removed from further
consideration for the next iteration. Thus misleading data is
eliminated and the building of a coherent nominal local model based
on the remaining consistent data is enabled therewith.
According to a further embodiment of the invention the
corresponding data of those turbines which have been removed as
outliers from further consideration within the global nominal
dataset are removed also from further consideration within the
respective clustered test data set accordingly. It can be expected,
that those data, which are not consistent within a fault free
reference period are also not consistent within a monitoring
period. Thus removing those data from consideration also from the
test dataset will improve the accuracy of the confidence factor
determined therefrom.
According to a further embodiment of the invention the at least two
same monitoring variables for each turbine (122, 124) are:
electrical measurements (e.g. generated electrical power, voltage,
current, power factor . . . ),
temperature measurements (e.g. nacelle temperature, electrical
generator temperature . . . ),
motional measurements (e.g. blade speed, electrical generator speed
. . . ) and/or
measurements variables describing the ambient conditions (e.g. wind
direction, wind speed and ambient temperature).
Those variables are easily to measure and in most cases available
in an existing SCADA system anyhow.
According to a further embodiment of the invention the frame data
of the turbines within the global nominal dataset comprise data
about the spatial proximity each to each other and/or the type of
the turbines. Those frame data are in important base for the
nominal global model based on the global nominal dataset which
describes the relationship inbetween the windmill turbines and
clusters the turbines accordingly. Turbines which are located in a
spatial proximity are subject to have a similar behavior since they
probably are subject to similar force impact of the wind and
windmill turbines of the same type might be subject to a similar
behavior since they are identical or at least similar. Thus
clustering of the windmill turbines is facilitated therewith.
Even the global model shows (if any) clusters of identical windmill
turbines, the geographical location of each wind turbine is
therefore not necessarily required, although this information could
be used to validate the clustering. If one compares it to the
geographical map of the windmill farm and finds that (some of) the
clusters could be explained by the geographical proximity of the
corresponding turbines, it is a good indicator that the obtained
global model captures the spatial location related heterogeneity
between the turbines operation.
According to a further embodiment of the invention the continuous
reference monitoring data of the global nominal dataset and the
continuous test monitoring data of the test dataset are in essence
a respective three mode dataset comprising several process
variables (index J) of several turbines (index I) along several
time samples (index K). Thus the most important data are storable
in a three dimensional array. Optionally respective flags could be
foreseen, indicating for example the assignment of a turbine to a
respective cluster or indicating the respective data as outlier to
be removed from consideration.
According to a further embodiment of the invention the data of the
global nominal dataset and/or the test dataset are collected and
provided at least predominantly by a SCADA system. A SCADA system
is typically foreseen in a windmill farm anyhow, so the collection
of required data can be done therewith in an easy way.
According to another embodiment of the invention a computing device
with a respective software program module running thereon is
foreseen for automatically performing the steps of the method. A
computing device can be for example an industrial PC with keyboard
and monitor which is embedded in a SCADA system. Thus a fully
automated monitoring and indicating of a critical turbine is
enabled.
According to a further embodiment of the invention automatic fault
analysis is initiated upon indicating a turbine as critical. Thus
it is further automatically evaluated, whether a critical turbine
is faulty respectively why it is indicated as critical so that
respective counteractions can be initiated.
According to a further embodiment of the invention the automatic
fault analysis comprises the following steps:
turbine level parsing,
time level parsing,
variable level parsing.
In the turbine level parsing it is determined, whether the
nonconformity index (NC) of any turbine exceeds a certain limit so
that the respective turbine is critical therewith. In subsequent
step the time level parsing the history of the NC of the respective
turbine is analyzed and the moment in that the NC exceeded the
certain limit is determined. Afterwards it is analyzed variable by
variable, whether there are irregularities at the moment determined
in the step before. This variable is typically a base for
identifying the root cause of a fault.
FIG. 1 shows the exemplary steps of the proposed workflow in a
sketch 10. The steps of the workflow are:
Nominal global model building,
Nominal local model building,
Test data projection and
Fault diagnostics on test dataset.
Nominal global model building
The data are collected from each windmill turbine measurement for
all windmill turbines present in the windmill farm to be monitored.
This data is first collected during a known fault-free time period
of operation and is preprocessed to form a nominal global dataset.
A global model is built using this global nominal dataset. This
global nominal model captures the relationship between all the
windmill turbines present in the windmill farm by statistical
techniques during a fault-free time period of operation. In order
to enhance the ability of the model to capture a deviation from
nominal behavior of a given windmill turbine, clusters of similar
windmill turbines are formed and the windmill turbines are divided
into groups according to obtained the clustering pattern. The
nominal global dataset is then accordingly divided into several
nominal local datasets. If there is no clear clustering pattern or
if the ability of the obtained global model to detect an abnormal
turbine behavior is considered as accurate enough, the global
nominal date set can also be used as a single nominal local
dataset.
The relationship between turbines in each nominal local dataset is
preferably modeled by the modeling algorithm described above, e.g.
preferably multivariate statistical algorithms. Outliers are
identified and removed from the nominal local dataset and the local
model is then rebuilt. The outlier removal/local model building
processes are iterated until no apparent outlier can be
identified.
The test dataset includes the same variables collected for the same
windmill turbines as the one used to build the nominal local
dataset. For the test dataset, data are collected during the time
period to be monitored and diagnosed. The test dataset is
pre-processed in a similar way as done for the nominal dataset. The
test dataset is projected on the nominal model. Projection here
refers to the operation of comparing the test dataset with the
nominal dataset by mean of using a NC index that quantifies the
nonconformity of test dataset to the nominal local model generated
from the nominal dataset. The NC index of the test dataset with
respect to the nominal local model is evaluated at each data
point.
The NC index values of the test dataset are parsed to provide the
condition of all windmill turbines, fault detection,
identification, isolation and process recovery. A fault here refers
to a component failure or a performance degradation of a single
windmill turbine.
FIG. 2 shows an exemplary nominal global model cluster in a sketch
20. The model results might exhibit clusters of windmill turbines
such as depicted in the sketch 20. The clustering pattern can be
the result of e.g. the spatial proximity of windmill turbines
leading to a similar wake effect affecting them and/or the fact
that some windmill turbines are of the same type. To improve the
accuracy of the model, windmill turbines can be divided into
several groups according to the obtained clustering pattern. The
dataset collected for each group of windmill turbines is used to
build a nominal local model in the next step of the workflow. If no
clear cluster is identified, the nominal global model is used as a
local model in the next step of the workflow.
FIG. 3 shows an exemplary flow chart of the "nominal global model
building" step in a sketch 30. The nominal model building step
includes the nominal dataset preprocessing, the global model
building, the windmill turbines clusters identification and the
local model building. The validity of the data is first verified in
order to identify potential erroneous data. According to the
results of the validation, erroneous data are removed. A global
model is built using the validated and preprocessed nominal dataset
to capture the relationship between the turbines. The term "global"
refers here to the fact that a single model includes all the
windmill turbines present in a given windmill farm.
FIG. 4 shows an exemplary flow chart of the "nominal local model
building" step in a sketch 40. While the global model is built to
identify groups of similar windmill turbines in a windmill farm,
the local model is built to model the similarity of windmill
turbines in the same group by means of the modeling algorithm.
Outliers are identified based on the results of the obtained local
model. A NC index and its confidence limit may be used to help
identifying an outlier. The identified outlier is removed from the
nominal local dataset and a new local model is built. The outlier
identification/model building processes are iterated until no
apparent outliers can be identified or until a set level of
homogeneity among the windmill turbines in the group is reached.
The nominal local model is used as a reference for the monitoring
and fault diagnosis in the subsequent steps.
FIG. 5 shows an exemplary flow chart of the "test data projection"
step in a sketch 50. The test dataset includes the measurements of
the same variables from the same windmill turbines present in the
nominal dataset and measured during the monitoring period. The test
dataset is divided into the same groups as the one used for the
nominal dataset in Step 1. In each group the outliers identified in
the nominal local dataset are removed from the test dataset so that
the windmill turbines in each test local dataset are the same as in
the corresponding nominal local dataset. As a result, the qth
nominal local model can be used to diagnose the condition of the
qth test local dataset by projecting the test local dataset on the
nominal local model. Depending on the modeling algorithm, the
realization of the projection operation can be different. The
operation `Projection` here means a conformity check between the
windmill turbines characteristics captured by the nominal local
model using the nominal local dataset and the characteristics of
the turbines present in the test local dataset. The nonconformity
is measured by the NC index at each data point of the ith windmill
turbine, jth variable and kth time point. The qth test local
dataset is then projected into the qth nominal local model. The NC
index is evaluated for each variable of each turbine at each time
point. The NC index is used for fault diagnostic in the next
step.
FIG. 6 shows an exemplary fault diagnostics on test monitoring data
in a sketch 6. Fault diagnostics includes three tasks:
fault detection,
fault isolation and
fault identification.
When possible, fault diagnostics can also provide the user a
support for a corrective action selection for a subsequent process
recovery. The proposed method is an integrated method which
achieves the four tasks using a single nominal model and parsing
the NC indices level by level.
FIG. 7 shows an exemplary turbine level parsing (level I) in a
sketch 70. In this level, the NC index is evaluated for each
turbine over the duration of the test dataset. The NC indices
computed for each windmill turbine are compared to each other
and/or with the confidence limit, as shown in the sketch 80.
Windmill turbines with higher NC values than the confidence limit
are flagged as faulty. Each of the faulty windmill turbines is
further analyzed in the next level.
FIG. 8 shows an exemplary time level parsing (level II) in a sketch
80. In this level, the NC index evaluated for each faulty windmill
turbine detected in level I is parsed along the time so that the
time trend of the fault of the turbine can be analyzed. The time
point or the time interval when the when the faulty windmill
turbine should be flagged as faulty can be identified using a
computed confidence limit.
FIG. 9 shows an exemplary variable level parsing (level III):
faulty variable isolation using statistical confidence limit in a
sketch 90. The NC index evaluated at the time when the turbine is
flagged as faulty are then parsed over all the variables. The
contribution of each variable to the total NC index value at this
time point can be compared to a confidence limit as shown in the
sketch 90.
FIG. 10 shows an exemplary variable level parsing (level III):
faulty variable isolation using comparison to contribution during
nominal period in a sketch 100. The comparison can also be made
using a contribution plot computed using data collected when the
turbine operates normally, the variables which are abnormally
contributing to the sum of the NC index values can therefore be
isolated.
FIG. 11 shows an exemplary signal level display (level IV) in a
sketch 1 10, Through level I to level III, the whole process of
fault detection, fault identification and isolation (fault
diagnosis) are achieved. The time trend of the isolated faulty
variable{circumflex over ( )}) of the faulty turbine(s) are plotted
along the time together with the expected time trend of the same
variable(s), i.e. the trend of this variable when it has a normal
level of contribution to the NC index value can also be simulated
and plotted against the actual variable time trend, as shown in
sketch 1 10. The deviation between the two time trends also offers
a direct and easily understandable visualization of the magnitude
of the fault. Moreover, the control system can be automatically or
manually adjusted based on the identified faulty component and the
expected value of the variable so that a process recovery can be
achieved.
FIG. 12 shows a schematic representation of a method for monitoring
various clusters of turbines of a windmill farm, including first
cluster 126, second cluster 128, and third cluster 130. First
cluster 126 includes first turbine 122 and second turbine 124.
Clusters 126, 128, 130 are connected to computing device with
monitoring system 132, which is in turn connected to database with
global nominal dataset 134 and display device 136.
While the invention has been illustrated and described in detail in
the drawings and foregoing description, such illustration and
description are to be considered illustrative or exemplary and not
restrictive. It will be understood that changes and modifications
may be made by those of ordinary skill within the scope of the
following claims. In particular, the present invention covers
further embodiments with any combination of features from different
embodiments described above and below. Additionally, statements
made herein characterizing the invention refer to an embodiment of
the invention and not necessarily all embodiments.
The terms used in the claims should be construed to have the
broadest reasonable interpretation consistent with the foregoing
description. For example, the use of the article "a" or "the" in
introducing an element should not be interpreted as being exclusive
of a plurality of elements. Likewise, the recitation of "or" should
be interpreted as being inclusive, such that the recitation of "A
or B" is not exclusive of "A and B," unless it is clear from the
context or the foregoing description that only one of A and B is
intended. Further, the recitation of "at least one of A, B and C"
should be interpreted as one or more of a group of elements
consisting of A, B and C, and should not be interpreted as
requiring at least one of each of the listed elements A, B and C,
regardless of whether A, B and C are related as categories or
otherwise. Moreover, the recitation of "A, B and/or C" or "at least
one of A, B or C" should be interpreted as including any singular
entity from the listed elements, e.g., A, any subset from the
listed elements, e.g., A and B, or the entire list of elements A, B
and C.
LIST OF REFERENCE SIGNS
10 exemplary steps of the proposed workflow according to the
invention 20 exemplary nominal global model cluster 22 first
cluster 24 second cluster 26 third cluster 30 exemplary flow chart
of the "nominal global model building" step 40 exemplary flow chart
of the "nominal local model building" step 50 exemplary fault
diagnostics on test monitoring data 60 exemplary fault diagnostics
on test monitoring data 62 first step turbine level parsing 64
second step time level parsing 66 third step variable level parsing
68 fourth step signal level display 70 exemplary turbine level
parsing (level I) 72 confidence limit of nonconformity index (NC)
74 first faulty turbine 76 second faulty turbine 80 exemplary time
level parsing (level II) 82 confidence limit of nonconformity index
(NC) 84 start of alarm 90 exemplary variable level parsing (level
III): faulty variable isolation using statistical confidence limit
92 faulty variable 100 exemplary variable level parsing (level
III): faulty variable isolation using comparison to contribution
during nominal period 102 faulty variable 1 10 exemplary signal
level display (level IV) 1 12 real signal 1 14 expected signal 120
exemplary windmill farm 122 first turbine of windmill farm 124
second turbine of windmill farm 126 first cluster of windmill farm
128 second cluster of windmill farm 130 third cluster of windmill
farm 132 computing device with monitoring system 134 database with
global nominal dataset 136 display device
* * * * *